A study on visual feature extraction in compressed domain for MPEG-7 descriptor

Abstract

MPEG-7 descriptors are known as standard descriptors for the content-based indexing and retrieval. Extracting visual features in compressed domain could lead to fast generation of MPEG-7 database as well as fast retrieval. In this paper, we propose fast feature extraction methods that can be applied to MPEG-7 visual descriptor from DCT compressed image, which is popularly used in image standard format. Particularly, in the proposed method, the fast extraction of color descriptors is achieved by combination of down-sampling and direct calculation of spatial information in DCT domain. Further, we propose modified edge histogram for fast extraction. To evaluate the proposed methods, experiments were performed with MPEG-7 XM and reference image database. The results show that the proposed methods achieve the reduction of feature extraction time with same performance as the conventional methods in image domain. Therefore, it is believed that the proposed methods can be applied to fast feature extraction with large size of database, thereby achieving fast retrieval.